n8n Review 2026

n8n (pronounced "n-eight-n") started as the hacker's answer to Zapier: a visual workflow automation tool you could self-host, extend with code, and run without per-task pricing anxiety. Then the AI wave arrived, and n8n turned out to be sitting on exactly the right substrate — workflows are where agents live. Its AI agent nodes, which plug large language models into tools, memory and multi-step workflows, made it a widely used platform for building real, working AI agents in 2025 and 2026.

These rankings are research-based — compiled from documentation, pricing pages and broad community consensus rather than hands-on lab benchmarks — and our independent testing program is still underway; test notes will be added to each pick as results come in. Read how we rank agents for the full process.

This review covers what n8n does in 2026, what self-hosting versus cloud really means for you, and the honest tradeoff at its core: power and control in exchange for a learning curve. It anchors our automation agents roundup.

Overview

At its core, n8n is a visual workflow builder: you connect nodes — triggers, app integrations, logic, code, AI models — into flows that run when something happens. A new row in a spreadsheet can trigger enrichment via an API, a draft written by an LLM, and a message posted to Slack. Hundreds of built-in integrations cover the common SaaS tools, and anything with an API is reachable through HTTP nodes.

The AI layer is what elevated n8n from automation tool to agent platform. AI agent nodes let you give a language model a goal, a set of tools (which can be other n8n workflows) and memory — and let it work. Builders use this for support agents that read the docs and the ticket queue, research agents that gather and synthesize, and internal copilots grounded in company data. Because the agent runs inside your workflows, every step is visible, loggable and debuggable — a meaningful contrast with black-box agents.

The self-hosting story remains the soul of the product. Run n8n on your own server and your data stays within your infrastructure as long as you also self-host your models — the moment your workflows call an external model API (OpenAI, Anthropic or similar), that data is sent to the provider. Your costs are your server bill plus operations labor and any external model API fees, with no per-execution meter running. That proposition — control plus economics — is why it has such a devoted technical following, and why it keeps showing up wherever serious agent builders gather.

What n8n-built agents look like in the wild tells you what the platform is for. A support agent that reads the docs, checks the ticket queue and drafts replies for human approval. A research agent that monitors sources, synthesizes findings and posts a digest to Slack every morning. A sales agent that enriches new leads, scores them and routes the hot ones to a human. The pattern is consistent: the LLM provides judgment and language, while n8n provides the tools, the memory, the scheduling and the audit trail. Builders choose n8n when they want agents they can inspect — every step visible in the workflow canvas, every run logged.

The honest costs are time and responsibility. n8n's learning curve is real: triggers, nodes, expressions, credentials, error handling — the concepts are simple individually but compose into genuine engineering work. Self-hosting adds the operational load: updates, backups, uptime, security patches. Teams routinely underestimate both, and the community forums are full of ambitious first projects that stalled on webhook configuration. Start small — one workflow that saves an hour a week — and let the platform earn its complexity before you build the agent swarm.

Getting started is easier than the reputation suggests, thanks to two things: a large library of community templates — prebuilt workflows for common jobs you can import and adapt — and Docker-based self-hosting that gets a personal instance running in minutes for anyone comfortable with a terminal. The sensible on-ramp is a template close to your need, modified rather than built from scratch. And n8n's community is genuinely active: forum answers, shared workflows and public examples mean most problems you hit have been solved by someone already, which shortens the learning curve considerably.

The first workflow is the hardest and the most instructive: pick one annoying recurring task, find the closest community template, and adapt it until it runs on a schedule. You will learn triggers, credentials and error handling in a single afternoon — and you will have something that saves you time every week. That is the n8n flywheel: each workflow teaches the skills for the next, and the library of automations becomes a genuine asset.

A concrete example: lead triage, end to end

Abstract descriptions only go so far. Here is what an n8n workflow actually looks like — a lead-triage pipeline of the kind small B2B teams run (illustrative, assembled from documented node types and common community patterns):

1. Webhook trigger. A new lead submits the website form; the form posts to an n8n webhook URL.
2. Enrichment. An HTTP node calls an enrichment API with the company domain and appends firmographic data — headcount, industry, tech stack.
3. AI scoring. An AI agent node reads the enriched lead plus your ideal-customer notes and returns a score with a one-line reason. If this step calls a pay-per-use model, it incurs token fees on every run — billing varies by node type and model.
4. Router. An IF node splits the flow: hot leads go one way, everyone else another.
5. Action. Hot leads create a CRM record, assign an owner, and post to a #hot-leads Slack channel. The rest get a personalized follow-up email drafted by the LLM and queued for human review.
6. Logging. Every execution is logged with inputs and outputs — so when a lead is scored oddly, you can open the run and see exactly which node decided what.

Nothing here requires exotic infrastructure: six nodes, two API credentials, one prompt. The pattern generalizes — swap "lead" for "support ticket" or "invoice" and the shape is the same. This is why n8n compounds: the second workflow reuses the credentials, the error handling, and the logging discipline you built for the first.

Key features

• Visual workflow builder. Drag-and-drop canvas for connecting triggers, integrations, logic and code into automations you can see and reason about.
• AI agent nodes. Give LLMs tools, memory and goals inside your workflows to build agents grounded in your own systems and data.
• Self-hosting. Run n8n on your own infrastructure under its fair-code license — your data stays under your control (unless your workflows call external model APIs), with no per-execution metering.
• 400+ integrations. A large library of prebuilt app connectors, plus generic HTTP, webhook and database nodes for everything else.
• Code nodes. Drop into JavaScript or Python anywhere in a workflow for logic the visual nodes cannot express.
• LangChain-based AI plumbing. Deep integration with the LangChain ecosystem for chains, tools, memory and model routing.
• Cloud option. A managed hosted version for teams that want n8n's power without operating servers.

Pricing

As of October 2026: self-hosting n8n is free under its fair-code license. Managed cloud plans start at €20/mo, billed annually (Starter) (n8n.io/pricing, verified Oct 8, 2026), with higher tiers for heavier workflow volumes.

The economics are the pitch. Self-hosted, your cost is a server — a few dollars a month of VPS handles a surprising amount of automation — versus per-task pricing that scales with your success on hosted competitors. The cloud plans exist for teams that would rather pay n8n to operate the infrastructure; they are competitively priced against the no-code incumbents. Either way, check the current tier details on the official site, since execution allowances and feature gates evolve.

Prices change frequently — confirm on the official site before buying.

The cost comparison that sells n8n is against per-task automation pricing at scale. A business running tens of thousands of automation tasks a month can pay hundreds on hosted per-task platforms; the same workload self-hosted on n8n costs a small VPS and some setup time. Even the cloud plans compare favorably once volume grows. The break-even analysis should include your time, though: if you value engineering hours highly and your automation needs are simple, a pricier but simpler tool can be the rational choice. n8n wins when volume is high, logic is complex, or control is non-negotiable.

The costs people forget

n8n's headline economics — free to self-host, flat pricing on cloud — are real. The costs below are the ones that surprise people anyway:

• Model API fees (self-hosted or cloud). The big one. n8n doesn't charge per execution when self-hosted, but steps that call pay-per-use models still incur token fees. In high-frequency or long-context workflows, model fees can exceed the platform subscription. Budget the model, not just the platform.
• Operations labor. Updates, backups, uptime monitoring, SSL certificates, debugging failed executions at odd hours. None of this appears on a pricing page. For a side project it's a hobby; for a business workflow it's a real line item — value your time honestly when comparing against managed alternatives.
• The learning curve as calendar time. Expect your first serious workflow to take days, not hours. Templates shorten this, but expressions, credentials, and error handling are genuinely new concepts for most people.
• Cloud execution burn. On cloud plans, watch execution allowances and active-workflow limits on lower tiers. A workflow that polls every minute creates about 43,200 polling opportunities in 30 days — how many count as billed executions depends on the trigger type and the platform's metering rules. Check your tier covers your actual pattern before you build twenty of them.
• Third-party APIs inside the workflow. The lead-triage example above calls an enrichment API; those services bill per lookup. n8n orchestrates for free — the services it orchestrates don't.

None of these are reasons not to use n8n. They're reasons to budget the system, not just the software — the mistake is comparing n8n's €0 self-hosted license against a competitor's $50 plan and calling the difference pure savings.

Who it's best for

n8n is best for technical builders — developers, ops engineers, technical founders and automation-minded teams — who want to construct AI agents and automations they fully control. If you are comfortable with APIs, webhooks and a little JavaScript, n8n gives you more power per dollar than most alternatives in the automation space, and the self-hosting option goes a long way on the data-sovereignty question.

It is not for non-technical users who want automation without thinking about infrastructure — they will be happier with a simpler no-code tool. And if you want a finished agent rather than a platform for building agents, look at the products in our best AI agents overview or the done-for-you options in our free AI agents roundup. n8n is the workshop, not the product off the shelf.

Who should skip it? Non-technical users who want automation to ‘just work’ will find n8n's concepts — webhooks, JSON expressions, API credentials — an unwelcome education; simpler no-code tools exist for them. Teams without anyone comfortable operating infrastructure should use the cloud plans rather than self-hosting, or they will learn ops the hard way. And if you need a finished AI agent rather than the means to build one, buy the agent — the workshop is only worth it if you enjoy building.

The bottom line

n8n is the agent builder's workshop: the rare platform that is simultaneously one of the cheapest serious options (self-hosted), one of the most transparent (every step visible), and among the most powerful (real code, real logic, real AI agents inside workflows). The price of all that is the learning curve and the operational responsibility of self-hosting. If you build agents rather than just using them, n8n belongs in your stack — and if you do not self-host, the cloud plans are still fairly priced for what you get.

Check current cloud tiers and self-hosting docs on the official site.

n8n's position keeps strengthening as agents go mainstream, because every agent eventually needs what n8n provides: tools to act through, memory to persist, schedules to run on, and logs to debug with. The chatbots get the headlines, but the workflows do the work — and n8n is where a remarkable share of that work gets built. For technical teams, it is less a question of whether to adopt it than when the first workflow pays for the learning curve. Start with the hour-a-week win; the agent platform reveals itself from there.